How people use Copilot for Health

2025-06-26 · Source: cs.AI updates on arXiv.org · Field: Health & Wellbeing — Medical Devices & Health Technology, Healthcare Systems & Policy, Artificial Intelligence & Machine Learning · Depth: Advanced, extended

Summary

An analysis of over 500,000 de-identified health-related conversations with Microsoft Copilot from January 2026 reveals key patterns in how people use conversational AI for health. Researchers developed a hierarchical intent taxonomy with 12 primary categories, validated by expert human annotation, and applied LLM-driven topic clustering. Findings indicate that nearly one in five conversations involve personal symptom assessment or condition discussion, with general information (40%) also concentrating on specific treatments. One in seven personal health queries are for someone other than the user, suggesting a caregiving role for AI. Personal queries and emotional health topics increase significantly in the evening and nighttime. Mobile device usage focuses on personal health, while desktop use is for professional and academic work. A substantial portion of queries also addresses healthcare system navigation, such as finding providers or understanding insurance.

Key takeaway

For research scientists developing health AI, understanding these usage patterns is crucial for responsible design and deployment. You should focus on tailoring AI responses based on device context and time of day, recognizing the prevalence of personal health and caregiving queries, especially during off-hours. This insight helps ensure appropriate information delivery and timely redirection to professional care, enhancing user safety and system utility.

Key insights

Conversational AI serves diverse health needs, from personal symptom checks to caregiving and system navigation, with usage patterns varying by device and time.

Principles

• Device choice signals user intent.
• AI acts as a caregiving tool.
• Usage peaks when traditional care is limited.

Method

A hierarchical intent taxonomy of 12 categories was developed using privacy-preserving LLM-based classification, validated by human annotation, and applied to de-identified conversation summaries for topic clustering.

In practice

• Design platform-specific health AI experiences.
• Prioritize safety for personal health intents.
• Address healthcare navigation friction.

Topics

• Microsoft Copilot Usage
• Health AI Intent Taxonomy
• LLM-driven Health Analysis
• Personal Health Concerns
• Healthcare System Navigation

Best for: Research Scientist, AI Scientist, AI Product Manager, AI Ethicist

Related on AIssential

• Microsoft debuts Copilot Health to unify medical records and fitness data — Copilot Health unifies disparate health data using AI to provide contextual narratives for users and doctors.
• A Taxonomy of Mental Health and Technology Needs for Alzheimer's and Dementia Caregivers — A new taxonomy links AD/ADRD caregiver mental health needs with technology interventions, addressing research gaps and guiding design.
• Calibre emerges from stealth with $3.3M to tackle “health guesswork” through causal AI — Causal AI combined with clinical expertise can deliver personalized, proactive health navigation at an accessible price point.
• Quoting Chengpeng Mou — ChatGPT sees substantial healthcare-related usage, especially outside clinic hours and in underserved areas.
• Talking to Your Data: Exploring Embodied Conversation as an Interface for Personal Health Reflection — Embodied conversational agents can facilitate objective personal health data reflection, moving beyond passive dashboard viewing.
• Lets use AI to help people manage illness — AI's true impact in healthcare lies in novel applications for illness management, not just replicating existing wellness tools.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.